[Bug c/24542] potential integer overflow should be warned on assignment to wider variable

["zhangboyang.id at gmail dot com" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=24542

--- Comment #12 from Zhang Boyang <zhangboyang.id at gmail dot com> ---
Hi,

Sorry for filled a duplicate bug. But I'd like to suggest reconsider this
feature request. Here are two reasons:

1) "u64 = 1 << u32", "u64 = u32 * u32" are common mistakes in beginners, 

2) These expressions may introduce vulnerability especially on now-widely-used
64-bit machines:
  On a typical 64-bit machine, it's ok to write:
    unsigned x = ...;
    malloc(sizeof(...) + x)
  but it will introduce vulnerability with a trivial change of "*2", i.e.:
    malloc(sizeof(...) + x * 2)
If expression is very long, it's very hard to find out where is the bug.

Instead of warn on multiplys, I suggest a new "-Wexpr-conversion", it will
detect and warn on implicit conversions if and only if: 1) convert to wider
variable, and 2) value is real expression (i.e. result of operands, like a*b;
but not variable or function call or explicit cast)

For example, it should warn on:

  uint64_t u64 = ...;
  uint32_t u32 = ...;
  u64 = 1 << u32;
    //  ^^^^^^^^
    //   suggests "u64 = (uint64_t)1 << (uint64_t)u32"
    //   suppressed by "u64 = (uint32_t)(1 << u32)"

But not on:
  u64 = u32;
  u64 = (u32)(...);
  u64 = f(...);

This might be a kind of noisy warning like "-Wconversion" but I believe it will
help some people (we can just disable it by default).

Zhang Boyang